Richard Bassed Yasmeen George Jiangning Song Beena Kumar Hans Schneider James Doery ZongYuan 宗元 Ge 戈 #digital #pathology
A new study addresses the limitations of current #ArtificialIntelligence models in diagnosing less frequent diseases using imaging data. By using a deep anomaly detection (AD) approach trained only on common diseases, the researchers aimed to detect infrequent pathologies in gastrointestinal biopsies. They trained their model on 17 million histological images from 5423 cases, achieving high accuracy in detecting both common and rare diseases, including cancers, with an area under the receiver operating characteristic curve of up to 97.7%. The model effectively generalizes across scanners and hospitals and highlights anomalous areas to assist pathologists in diagnostics. This novel AD technique could improve AI safety and adoption in clinical histopathology by detecting rare diseases, including cancers, without specific training. Read the Original Article “AI-Based Anomaly Detection for Clinical-Grade Histopathological Diagnostics” by J. Dippel et al.: https://nejm.ai/3YwvUZz #AIinMedicine
